The Verified Number: Golf, Data Centers, and Water
American golf courses use 4.6 times more water than every data center in the country combined. That number checks out. But it stacks today against tomorrow, and the water you can see against the water nobody's counting.
The Verified Number: Golf, Data Centers, and Water
4.6 times. That's the ratio between how much water American golf courses use and how much every data center in the country uses combined. Golf allegedly pumps 2.08 billion gallons a day; data centers, 449 million.
Kevin O'Leary, the Shark Tank investor, dropped that number on Glenn Beck's radio show while defending his own data center project in Utah. His line: "any data center today is basically a golf course."
The number holds up. We went and checked it, and golf genuinely does drink far more water than servers, today, nationwide. The problem isn't the arithmetic. It's what that arithmetic is quietly leaving out.
A twist on the format: distortion by framing
This is the second edition of this series. The premise is simple: the number is almost always accurate, it's the story wrapped around it that goes sideways. Last week, with the Boston College study on seniors, the distortion lived in the headlines, not the study itself.
This time it lives somewhere else: in the framing of a spokesperson who has a direct stake in getting people to buy his reading of the data. This isn't an indictment. Waving a real statistic around to defuse a concern is standard practice, in politics and in business alike. What matters here is the mechanism, not the man: how a perfectly accurate number can tell a false story once you carefully choose what goes into the comparison.
Where the number comes from
Both figures are real. The golf number traces back to GCSAA data, the trade group for course superintendents. The data center figure comes from a Florida water utilities association, and was picked up by The Next Web, which fact-checked O'Leary's claim. TNW's own conclusion: on the substance, he's not wrong to point it out.
Still, the outlet calls the argument "deflection dressed up as statistics." Translation: a solid answer to a question nobody asked. The real question isn't which one, golf or servers, drinks more water today. It's where both curves are heading, and what the number leaves out of frame.
Omission number one: time
A single number is a snapshot. O'Leary presents it as a permanent truth. But the two curves aren't pointing the same way.
On the golf side, water use is dropping. Per GCSAA, American courses used 1.63 million acre-feet in 2024, about 531 billion gallons for the year, 31% less than in 2005. Golf spent twenty years installing precision sprinklers, drought-tolerant turf, and recycled irrigation water. Its consumption is flat, if not declining.
Data centers are moving the opposite direction. The 2024 Lawrence Berkeley National Laboratory report, the field's reference study, puts direct cooling water at 17.4 billion gallons in 2023, and projects 38 to 73 billion gallons by 2028. That's usage doubling or quadrupling in five years.
TNW places the crossover somewhere around 2026 or 2027. The exact year is debatable depending on what's counted, but the direction isn't: one flat line, one climbing fast. Comparing their heights on a random Tuesday in 2026 says nothing about where they're going.
Omission number two: the water you don't see
Here's the part that carries the most weight, and gets buried the most. O'Leary's number only counts direct water: what evaporates in the cooling towers, on site, at the data center. That's the visible water, the kind that shows up on the building's meter.
But a data center's biggest draw is electricity, and generating that electricity uses water too, at the power plant, often hundreds of miles away. The same LBNL report estimates that indirect water at roughly 211 billion gallons in 2023, about twelve times the direct cooling total. It's real water, pulled from real watersheds, it just doesn't appear anywhere near where O'Leary is pointing.
Count both, and the comparison shifts entirely. This isn't a methodology footnote. It's the whole story.
We've seen this pattern before: data centers' green metrics fall apart once you actually recalculate them. Water follows the same logic. You measure what's convenient, and call it a scorecard.
Even the golf number is the biggest one available
One last detail seals the framing. The 2.08 billion gallons a day attributed to golf is an old estimate, from before the industry's big water-saving push. GCSAA now puts golf's usage closer to 1.45 billion gallons a day. On one side of the scale, O'Leary reaches for the largest number in circulation. On the other, the narrowest. Both choices push in the same direction.
What the number is actually being used to say
Let's separate what the number says from what it's being made to say. What it says: nationwide, golf currently pumps more direct water than servers do. True. What it's made to say: therefore, a data center isn't a water problem. False, and beside the point.
A national average tells you nothing about one specific plant on one specific watershed. O'Leary's Utah project calls for a first phase of 1.4 gigawatts, and Utah Clean Energy estimates the site could use between 2 and 16.6 billion gallons a year depending on its power source. The local question stays wide open: can that particular aquifer actually spare that particular volume. The golf course down the road doesn't answer it.
That's the mechanism. Take an accurate number, shrink its scope, freeze its timeline, and it ends up propping up a conclusion its own arithmetic can't support. The number isn't lying. It's the frame around the number that tells the story, and the frame is what's worth reading.
